ECD-FineGrained-Conifer

该数据集是由ECD-FineGrained和ECD-Conifer数据集合并而成，并进一步通过UltraFeedback进行了增强。此外，为了提升数据质量，该数据集还移除了那些在GPT-4评分中得分相同的偏好配对。该数据集的任务是用于生成偏好数据，以改善对指令遵循能力的增强。This dataset is compiled by merging the ECD-FineGrained and ECD-Conifer datasets, and further augmented via UltraFeedback. Additionally, to improve data quality, all preference pairs that obtained identical scores in GPT-4 evaluations have been removed from this dataset. The task of this dataset is to generate preference data for optimizing the enhancement of instruction-following capabilities.